Brain connectivity fingerprinting and behavioural prediction rest on distinct functional systems of the human connectome

Abstract

The prediction of inter-individual behavioural differences from neuroimaging data is a rapidly evolving field of research focusing on individualised methods to describe human brain organisation on the single-subject level. One method that harnesses such individual signatures is functional connectome fingerprinting, which can reliably identify individuals from large study populations. However, the precise relationship between functional signatures underlying fingerprinting and behavioural prediction remains unclear. Expanding on previous reports, here we systematically investigate the link between discrimination and prediction on different levels of brain network organisation (individual connections, network interactions, topographical organisation, and connection variability). Our analysis revealed a substantial divergence between discriminatory and predictive connectivity signatures on all levels of network organisation. Across different brain parcellations, thresholds, and prediction algorithms, we find discriminatory connections in higher-order multimodal association cortices, while neural correlates of behaviour display more variable distributions. Furthermore, we find the standard deviation of connections between participants to be significantly higher in fingerprinting than in prediction, making inter-individual connection variability a possible separating marker. These results demonstrate that participant identification and behavioural prediction involve highly distinct functional systems of the human connectome. The present study thus calls into question the direct functional relevance of connectome fingerprints.

Fig. 1. Within-network distribution of selected edges in fingerprinting and behavioural prediction.

a Visualises the percentage of selected edges for fingerprinting and for prediction within each network, adjusted for the total number of edges in each network. b–d shows the prediction results of three psychometric variables of interest. Language comprehension was evaluated using the picture vocabulary task.

Fig. 2. Single-edge and between-network overlap for fingerprinting and prediction.

a Grey lines mark highly discriminatory edges for fingerprints and predictive edges for behaviour, thresholded at the 99th percentile or p < 0.01 respectively. Red lines, only available for strength, visualise overlap. For fluid intelligence and language comprehension, no edges overlapped. b Shows the entire network-by-network matrix of the same selected edges, adjusted for a total number of edges.

Fig. 3. Spatial distribution of node degrees.

Distribution of node degrees for discriminatory nodes a and behaviourally predictive nodes (b–d) on the left. The edges underlying the node degrees are thresholded at the 99th percentile and p < 0.01 respectively. The right-hand side displays the spin permutation results, with red lines marking the empirical correlation of discriminatory and predictive nodes.

Fig. 4. Overlap of discriminatory edges and high-variability edges.

a Black lines designate overlapping edges between the top one, two or five percent of discriminatory edges and high-variability edges. Grey lines depict non-overlapping discriminatory edges. b Shows the distribution of edge standard deviation across participants. In the boxplot, the middle line signifies the median, the lower and upper hinges correspond to the first and third quartiles and the upper and lower whisker represent 1.5 times the respective interquartile ranges. All points outside of whiskers are outliers. ***p < 0.001, FDR-corrected.

Fig. 5. Control analyses for different functional atlases.

Results for Brainnetome, HCP MMP 1.0, and AAL atlases (left-to-right). From top to bottom, panels visualise a within and across-network connections, b spatial topology and spin permutation of nodes for fluid intelligence and fingerprinting, c the overlap of individual edges between highly discriminatory and high-variability edges, and d the distribution of edge standard deviations over participants in fingerprinting and behavioural prediction. In the boxplots, the middle line signifies the median, the lower and upper hinges correspond to the first and third quartiles and the upper and lower whisker represent 1.5 times the respective interquartile ranges. All points outside of whiskers are outliers. ***p < 0.001, FDR corrected.